training_log.json

{
  "base_model_name": "Undi95_Meta-Llama-3-8B-hf",
  "base_model_class": "LlamaForCausalLM",
  "base_loaded_in_4bit": true,
  "base_loaded_in_8bit": false,
  "projections": "q, v",
  "loss": 3.2517,
  "grad_norm": 0.5076477527618408,
  "learning_rate": 0.0,
  "epoch": 4.48,
  "current_steps": 627,
  "train_runtime": 6246.0484,
  "train_samples_per_second": 0.458,
  "train_steps_per_second": 0.006,
  "total_flos": 2.3779975492534272e+17,
  "train_loss": 3.460700845718384
}